News for hackers. Stuff that matters.
Re:Automation Should Be Like Iron Man, Not Ultron by Houshalter on Saturday May 21, 2016 @08:13PM (#11746834)
That works because computer game playing was totally different than how humans play. Humans can recognize common patterns, and computers can brute force many possible positions and evaluate them. The two complement each other.

But now AlphaGo beat humans using neural networks that can learn to recognize common patterns similar to humans. And it was combined with tree search to evaluate many more possible positions than a human could.

So I doubt human and AlphaGo would have any advantage over just AlphaGo. In the future the same will become true in many other domains.

Re:Automation Should Be Like Iron Man, Not Ultron by donlzx on Saturday May 21, 2016 @08:48PM (#11746936)
I actually think AlphaGo is a perfect example of Human-machine cooperation. Deep neural network is the easier part, the intuition and architecture design by Humans are the most critical part.
Re:Automation Should Be Like Iron Man, Not Ultron by barkingdog on Saturday May 21, 2016 @09:06PM (#11746990)
Here's where your analogy breaks down: what if the state of AlphaGo enhancers was as good as AlphaGo AIs? Could the current state of the art AI beat a human working with a state of the art augmented AI based assistant?

The lack of an answer is an interesting question, and I wonder how much it has to do with cultural concerns. Obviously, "machine beats man at intelligent task" is a lot sexier in some ways.

Re:Automation Should Be Like Iron Man, Not Ultron by rqebmm on Saturday May 21, 2016 @10:33PM (#11747165)
> I doubt human and AlphaGo would have any advantage over just AlphaGo

I know what you're saying, but I think it's important to remember that AlphaGo was created by humans. The nature of Go pattern recognition was almost assuredly factored into the creation of the algorithm.

Re:Automation Should Be Like Iron Man, Not Ultron by srtjstjsj on Sunday May 22, 2016 @12:32AM (#11747374)
No. AlphaGo used a general-purpose algorithm with basic Go rules and novice-strategy optimizations.